Universal AI Context Layer

The shared brain for
all your AI tools.

Your context is trapped inside each AI tool by design. Sulci captures it everywhere and surfaces it everywhere — so you stop re-explaining yourself.

without
with sulci
Sulci Cortex
0
Atoms
Links
0
Sources
0
Types
Live Atom Feed
Extract Classify Link Store

Two ways to run Sulci.
Pick what fits your setup.

Self-host locally with a one-liner. Your data stays on your machine. Or use Sulci Cloud: fully managed, zero setup, access from anywhere.

Sulci Cloud

No server needed

Fully managed. Sign up, connect your AI tools, and your knowledge is captured and synced. No local server, no API keys, no maintenance.

  • No installation or API keys required
  • Access from any device
  • Automatic updates & backups
  • 14-day Pro trial, free to start
  • Teams — the shared brain for your company
Get started free →

What happens after install

1
Cortex opens with the onboarding wizard at localhost:8080
2
Paste your API key — OpenAI or Anthropic — right in the wizard
3
Connect your tools — MCP for Claude/Cursor, Chrome extension, Slack, Calendar, Docs
Knowledge flows automatically — every AI conversation is captured, extracted, and shared

Every conversation
starts from zero.

You use multiple AI tools every day. Not one of them shares context with the others. Your history, preferences, and decisions stay locked inside each tool, by design. Your brain is the connective tissue holding your entire AI stack together.
Here's what that looks like in practice:

Without Sulci, you re-explained your setup four times today.
With Sulci? Zero. Same day. Same tools. Completely different experience.
01

Recall Drift

When you re-explain from memory, details mutate. Small errors compound. Claude debugs a system that doesn't exist, built from your inaccurate recall.

02

Contradictory Ground Truth

Decisions evolve, but each tool is frozen at the last thing you said. ChatGPT still thinks RabbitMQ while Claude plans your Kafka migration.

03

Confident Blind Spots

Every tool only sees its own conversations. It gives confident advice while missing critical context from every other tool you use.

🔒

This isn't an accident.

Shared context is portable. Portable means you can leave, and that kills the $20/month subscription. So your history, preferences, and decisions become an ever-growing switching cost. The fix won't come from the vendors. Your context should be yours.

Same week.
With Sulci.

Now every tool shares one brain. Ask anything and Sulci surfaces the right context instantly, regardless of which tool captured it.

Try a suggestion below

Claude Code forgets.
Sulci remembers.

Every Claude Code session starts with a blank slate. Your preferences, decisions, project context. Gone. Sulci gives Claude Code a persistent brain that survives across sessions and projects.

Without Sulci
$ claude
Starting new session...
You: Fix the auth bug in the Atlas API
Claude: I'd be happy to help! A few questions:
- What framework are you using?
- What's the auth strategy?
- Where are the auth files?
- What database?
You: We literally discussed all of this yesterday...
With Sulci
$ claude
Starting new session...
sulci://session-brief PostgreSQL 16 + pgvector FastAPI + JWT 15-min tokens Prefer async/await, type hints Sarah = tech lead, Marcus = devops
You: Fix the auth bug in the Atlas API
Claude: I see the Atlas project uses FastAPI with JWT (15-min access tokens, httpOnly cookies). Let me check the auth middleware...
Reading apps/api/src/auth/middleware.py
Found the bug: token refresh race condition on line 47
🔄

Cross-Session Memory

Preferences, decisions, and project context persist across every Claude Code session. No more re-explaining.

📂

Project-Aware

Auto-detects which project you're in. Atlas context stays with Atlas. Global preferences follow you everywhere.

🔗

Cross-Tool Knowledge

Decisions from ChatGPT, research from Claude, calendar events, Slack threads. All available in Claude Code.

Four stages.
Total recall.

01

Capture

Conversations flow in via Chrome extension, MCP server, API proxy, or REST API. Every interaction is captured automatically.

POST /api/v1/interactions
02

Extract

An LLM extracts structured knowledge atoms (facts, decisions, preferences, entities, relationships) with project scoping and conflict detection built in.

KnowledgeExtractor → gpt-4o-mini
03

Store

SQLite for relational data. ChromaDB for vector embeddings. Dual-indexed for both structured queries and semantic search.

SQLite + ChromaDB
04

Serve

Relevant context flows out via MCP tools, proxy injection, or API queries. Core atoms (importance ≥ 0.75) are always prepended first. Query-matched atoms follow, ranked by relevance, freshness, and confidence.

0.5×similarity + 0.2×freshness + 0.15×confidence + 0.15×frequency

Beyond CLAUDE.md

CLAUDE.md was a great start. But a static file in a single tool can't keep up
with how you actually work. Here's what changes when your context layer evolves.

🌐

Cross-Tool Knowledge

CLAUDE.md only works in Claude Code. Sulci works everywhere. Knowledge stored in one tool is instantly available in every other tool you use.

🔍

Vector Search, Not Token Dumping

Instead of dumping everything into every prompt, Sulci uses vector embeddings to surface only the most relevant context. Faster, cheaper, and more precise.

Living Knowledge Graph

CLAUDE.md is a static file you manually edit. Sulci builds a knowledge graph that understands relationships between atoms. It captures knowledge, detects contradictions, verifies facts, and evolves as your project evolves.

🔒

Security & Trust

Your knowledge is yours. Sulci runs 100% locally with AES-256 encryption at rest, automatic PII detection, scoped API keys with RBAC, and audit logging. Your data never leaves your machine.

Seven types of
structured knowledge.

Every piece of information extracted from your conversations is classified, structured, scored, and scoped to the right project.

Not all knowledge
is equal.

Sulci assigns every atom an importance score and semantic flags. The most critical facts — your architecture decisions, coding standards, hard constraints — are always in context. Everything else is retrieved only when relevant.

Core Knowledge — Always Present

Atoms with importance ≥ 0.75 are injected into every conversation automatically — before any query matching. Your architecture decisions, coding standards, and hard constraints are never missing, no matter what you ask.

🏷

Semantic Flags & Auto-Scoring

Tag any atom as CORE, DECISION, TECHNICAL, or PREFERENCE. Flags drive automatic importance scoring — instructions score 0.9 automatically, CORE-flagged atoms 0.9, decisions 0.85. No manual tuning required.

Temporal Validity

Facts can expire. Set a valid_until date and Sulci automatically stops injecting stale knowledge. Sprint goals, deadlines, and time-boxed decisions never pollute future context.

Atom Supersession

When a decision changes, record a new atom and mark it as superseding the old one. The old atom is expired automatically — no contradictions, no ghosts from the past. Your knowledge base evolves cleanly.

Plugs into
everything you already use.

Primary

MCP Server

Claude, OpenCode, Cursor, Windsurf & any MCP client

Native integration via Model Context Protocol. Any MCP-compatible coding assistant can query your knowledge, store new information, and extract from conversations through natural language. Auto-detects your current project from the working directory.

query_context add_knowledge record_interaction list_knowledge list_conflicts resolve_conflict verification_queue calendar_today slack_post_message gdocs_read
{
  "mcpServers": {
    "sulci": {
      "command": "/bin/sh",
      "args": ["-c", "uv --directory /path/to/sulci run --package sulci-mcp sulci-mcp"]
    }
  }
}
See demo →

Chrome Extension

ChatGPT, Claude, Gemini & Perplexity

Passive capture from web AI tools. Install once, and every ChatGPT, Claude, Gemini, and Perplexity conversation is automatically captured and extracted into knowledge atoms. Zero effort required.

ChatGPT Claude Gemini Perplexity Auto-capture
See demo → Chrome Web Store ↗

VS Code Extension

IDE Context Sidebar · Works in Cursor & VS Code

Context sidebar right in your editor. Auto-queries relevant knowledge as you navigate code. Type in the sidebar search box or hit Cmd+Alt+K for instant queries. Save any selection as a knowledge atom with Cmd+Alt+J.

Context Sidebar Save Selection Cmd+Alt+K Search Box Sulci Cloud
See demo → VS Code Marketplace ↗

Your data stays
yours.

Sulci is local-first and privacy-aware by design. No cloud sync, no data sharing, no surprises. You opt in to everything.

💻

Local by Default

All knowledge atoms live on your machine. SQLite database, local ChromaDB vectors. Nothing leaves localhost unless you explicitly enable sync mode for a project.

🔍

Automatic PII Detection

Every incoming atom is scanned for emails, phone numbers, SSNs, credit cards, and IP addresses. PII is blocked from standard projects. No accidental data leaks.

🔒

Private Projects

Need to store sensitive data? Create a Private project. It accepts PII, is clearly marked with a lock icon, and gets extra data handling protections throughout the system.

🎛️

Data Mode Control

Each project has its own data mode: Local (never leaves your machine) or Sync (eligible for cross-instance sharing). You control the boundary, per project.

Why I built
Sulci.

I use Claude Code, ChatGPT, Gemini, Copilot, and Claude every single day. And every single day, I re-explain the same context. "The project uses FastAPI." "I prefer async/await." "Marcus handles DevOps." Over and over.

None of these tools talk to each other. Decisions made in one are invisible to the rest. Context evaporates between sessions. I was losing hours every week to this friction.

So I built Sulci. A local-first knowledge layer that sits underneath all your AI tools, captures what matters, and makes it available everywhere, automatically. No cloud dependency. No data sharing. Your knowledge, on your machine, working for you.

Jaime López
Nuclear engineer turned growth technologist

14+ years in B2B tech across Wärtsilä, Aiven, and Ververica. Nuclear Energy Engineering from Technical University of Madrid. Statistics & Data Science from MIT. Based in Helsinki, building at the intersection of developer tools, AI, and revenue.

Give your AI tools a shared brain.

Sulci is in limited beta. Get early access and never re-explain yourself to an AI tool again.

or email directly: hello@lopez.fi